]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - fs/btrfs/disk-io.c
projects / mirror_ubuntu-artful-kernel.git / blob
? search:
summary | shortlog | log | commit | commitdiff | tree
blame | history | raw | HEAD
Btrfs: add a priority queue to the async thread helpers
[mirror_ubuntu-artful-kernel.git] / fs / btrfs / disk-io.c
1 /*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19 #include <linux/fs.h>
20 #include <linux/blkdev.h>
21 #include <linux/scatterlist.h>
22 #include <linux/swap.h>
23 #include <linux/radix-tree.h>
24 #include <linux/writeback.h>
25 #include <linux/buffer_head.h>
26 #include <linux/workqueue.h>
27 #include <linux/kthread.h>
28 #include <linux/freezer.h>
29 #include "compat.h"
30 #include "crc32c.h"
31 #include "ctree.h"
32 #include "disk-io.h"
33 #include "transaction.h"
34 #include "btrfs_inode.h"
35 #include "volumes.h"
36 #include "print-tree.h"
37 #include "async-thread.h"
38 #include "locking.h"
39 #include "ref-cache.h"
40 #include "tree-log.h"
41 #include "free-space-cache.h"
42
43 static struct extent_io_ops btree_extent_io_ops;
44 static void end_workqueue_fn(struct btrfs_work *work);
45
46 /*
47 * end_io_wq structs are used to do processing in task context when an IO is
48 * complete. This is used during reads to verify checksums, and it is used
49 * by writes to insert metadata for new file extents after IO is complete.
50 */
51 struct end_io_wq {
52 struct bio *bio;
53 bio_end_io_t *end_io;
54 void *private;
55 struct btrfs_fs_info *info;
56 int error;
57 int metadata;
58 struct list_head list;
59 struct btrfs_work work;
60 };
61
62 /*
63 * async submit bios are used to offload expensive checksumming
64 * onto the worker threads. They checksum file and metadata bios
65 * just before they are sent down the IO stack.
66 */
67 struct async_submit_bio {
68 struct inode *inode;
69 struct bio *bio;
70 struct list_head list;
71 extent_submit_bio_hook_t *submit_bio_start;
72 extent_submit_bio_hook_t *submit_bio_done;
73 int rw;
74 int mirror_num;
75 unsigned long bio_flags;
76 struct btrfs_work work;
77 };
78
79 /* These are used to set the lockdep class on the extent buffer locks.
80 * The class is set by the readpage_end_io_hook after the buffer has
81 * passed csum validation but before the pages are unlocked.
82 *
83 * The lockdep class is also set by btrfs_init_new_buffer on freshly
84 * allocated blocks.
85 *
86 * The class is based on the level in the tree block, which allows lockdep
87 * to know that lower nodes nest inside the locks of higher nodes.
88 *
89 * We also add a check to make sure the highest level of the tree is
90 * the same as our lockdep setup here. If BTRFS_MAX_LEVEL changes, this
91 * code needs update as well.
92 */
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
94 # if BTRFS_MAX_LEVEL != 8
95 # error
96 # endif
97 static struct lock_class_key btrfs_eb_class[BTRFS_MAX_LEVEL + 1];
98 static const char *btrfs_eb_name[BTRFS_MAX_LEVEL + 1] = {
99 /* leaf */
100 "btrfs-extent-00",
101 "btrfs-extent-01",
102 "btrfs-extent-02",
103 "btrfs-extent-03",
104 "btrfs-extent-04",
105 "btrfs-extent-05",
106 "btrfs-extent-06",
107 "btrfs-extent-07",
108 /* highest possible level */
109 "btrfs-extent-08",
110 };
111 #endif
112
113 /*
114 * extents on the btree inode are pretty simple, there's one extent
115 * that covers the entire device
116 */
117 static struct extent_map *btree_get_extent(struct inode *inode,
118 struct page *page, size_t page_offset, u64 start, u64 len,
119 int create)
120 {
121 struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
122 struct extent_map *em;
123 int ret;
124
125 spin_lock(&em_tree->lock);
126 em = lookup_extent_mapping(em_tree, start, len);
127 if (em) {
128 em->bdev =
129 BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
130 spin_unlock(&em_tree->lock);
131 goto out;
132 }
133 spin_unlock(&em_tree->lock);
134
135 em = alloc_extent_map(GFP_NOFS);
136 if (!em) {
137 em = ERR_PTR(-ENOMEM);
138 goto out;
139 }
140 em->start = 0;
141 em->len = (u64)-1;
142 em->block_len = (u64)-1;
143 em->block_start = 0;
144 em->bdev = BTRFS_I(inode)->root->fs_info->fs_devices->latest_bdev;
145
146 spin_lock(&em_tree->lock);
147 ret = add_extent_mapping(em_tree, em);
148 if (ret == -EEXIST) {
149 u64 failed_start = em->start;
150 u64 failed_len = em->len;
151
152 free_extent_map(em);
153 em = lookup_extent_mapping(em_tree, start, len);
154 if (em) {
155 ret = 0;
156 } else {
157 em = lookup_extent_mapping(em_tree, failed_start,
158 failed_len);
159 ret = -EIO;
160 }
161 } else if (ret) {
162 free_extent_map(em);
163 em = NULL;
164 }
165 spin_unlock(&em_tree->lock);
166
167 if (ret)
168 em = ERR_PTR(ret);
169 out:
170 return em;
171 }
172
173 u32 btrfs_csum_data(struct btrfs_root *root, char *data, u32 seed, size_t len)
174 {
175 return btrfs_crc32c(seed, data, len);
176 }
177
178 void btrfs_csum_final(u32 crc, char *result)
179 {
180 *(__le32 *)result = ~cpu_to_le32(crc);
181 }
182
183 /*
184 * compute the csum for a btree block, and either verify it or write it
185 * into the csum field of the block.
186 */
187 static int csum_tree_block(struct btrfs_root *root, struct extent_buffer *buf,
188 int verify)
189 {
190 u16 csum_size =
191 btrfs_super_csum_size(&root->fs_info->super_copy);
192 char *result = NULL;
193 unsigned long len;
194 unsigned long cur_len;
195 unsigned long offset = BTRFS_CSUM_SIZE;
196 char *map_token = NULL;
197 char *kaddr;
198 unsigned long map_start;
199 unsigned long map_len;
200 int err;
201 u32 crc = ~(u32)0;
202 unsigned long inline_result;
203
204 len = buf->len - offset;
205 while (len > 0) {
206 err = map_private_extent_buffer(buf, offset, 32,
207 &map_token, &kaddr,
208 &map_start, &map_len, KM_USER0);
209 if (err)
210 return 1;
211 cur_len = min(len, map_len - (offset - map_start));
212 crc = btrfs_csum_data(root, kaddr + offset - map_start,
213 crc, cur_len);
214 len -= cur_len;
215 offset += cur_len;
216 unmap_extent_buffer(buf, map_token, KM_USER0);
217 }
218 if (csum_size > sizeof(inline_result)) {
219 result = kzalloc(csum_size * sizeof(char), GFP_NOFS);
220 if (!result)
221 return 1;
222 } else {
223 result = (char *)&inline_result;
224 }
225
226 btrfs_csum_final(crc, result);
227
228 if (verify) {
229 if (memcmp_extent_buffer(buf, result, 0, csum_size)) {
230 u32 val;
231 u32 found = 0;
232 memcpy(&found, result, csum_size);
233
234 read_extent_buffer(buf, &val, 0, csum_size);
235 printk(KERN_INFO "btrfs: %s checksum verify failed "
236 "on %llu wanted %X found %X level %d\n",
237 root->fs_info->sb->s_id,
238 buf->start, val, found, btrfs_header_level(buf));
239 if (result != (char *)&inline_result)
240 kfree(result);
241 return 1;
242 }
243 } else {
244 write_extent_buffer(buf, result, 0, csum_size);
245 }
246 if (result != (char *)&inline_result)
247 kfree(result);
248 return 0;
249 }
250
251 /*
252 * we can't consider a given block up to date unless the transid of the
253 * block matches the transid in the parent node's pointer. This is how we
254 * detect blocks that either didn't get written at all or got written
255 * in the wrong place.
256 */
257 static int verify_parent_transid(struct extent_io_tree *io_tree,
258 struct extent_buffer *eb, u64 parent_transid)
259 {
260 int ret;
261
262 if (!parent_transid || btrfs_header_generation(eb) == parent_transid)
263 return 0;
264
265 lock_extent(io_tree, eb->start, eb->start + eb->len - 1, GFP_NOFS);
266 if (extent_buffer_uptodate(io_tree, eb) &&
267 btrfs_header_generation(eb) == parent_transid) {
268 ret = 0;
269 goto out;
270 }
271 printk("parent transid verify failed on %llu wanted %llu found %llu\n",
272 (unsigned long long)eb->start,
273 (unsigned long long)parent_transid,
274 (unsigned long long)btrfs_header_generation(eb));
275 ret = 1;
276 clear_extent_buffer_uptodate(io_tree, eb);
277 out:
278 unlock_extent(io_tree, eb->start, eb->start + eb->len - 1,
279 GFP_NOFS);
280 return ret;
281 }
282
283 /*
284 * helper to read a given tree block, doing retries as required when
285 * the checksums don't match and we have alternate mirrors to try.
286 */
287 static int btree_read_extent_buffer_pages(struct btrfs_root *root,
288 struct extent_buffer *eb,
289 u64 start, u64 parent_transid)
290 {
291 struct extent_io_tree *io_tree;
292 int ret;
293 int num_copies = 0;
294 int mirror_num = 0;
295
296 io_tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
297 while (1) {
298 ret = read_extent_buffer_pages(io_tree, eb, start, 1,
299 btree_get_extent, mirror_num);
300 if (!ret &&
301 !verify_parent_transid(io_tree, eb, parent_transid))
302 return ret;
303
304 num_copies = btrfs_num_copies(&root->fs_info->mapping_tree,
305 eb->start, eb->len);
306 if (num_copies == 1)
307 return ret;
308
309 mirror_num++;
310 if (mirror_num > num_copies)
311 return ret;
312 }
313 return -EIO;
314 }
315
316 /*
317 * checksum a dirty tree block before IO. This has extra checks to make sure
318 * we only fill in the checksum field in the first page of a multi-page block
319 */
320
321 static int csum_dirty_buffer(struct btrfs_root *root, struct page *page)
322 {
323 struct extent_io_tree *tree;
324 u64 start = (u64)page->index << PAGE_CACHE_SHIFT;
325 u64 found_start;
326 int found_level;
327 unsigned long len;
328 struct extent_buffer *eb;
329 int ret;
330
331 tree = &BTRFS_I(page->mapping->host)->io_tree;
332
333 if (page->private == EXTENT_PAGE_PRIVATE)
334 goto out;
335 if (!page->private)
336 goto out;
337 len = page->private >> 2;
338 WARN_ON(len == 0);
339
340 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
341 ret = btree_read_extent_buffer_pages(root, eb, start + PAGE_CACHE_SIZE,
342 btrfs_header_generation(eb));
343 BUG_ON(ret);
344 found_start = btrfs_header_bytenr(eb);
345 if (found_start != start) {
346 WARN_ON(1);
347 goto err;
348 }
349 if (eb->first_page != page) {
350 WARN_ON(1);
351 goto err;
352 }
353 if (!PageUptodate(page)) {
354 WARN_ON(1);
355 goto err;
356 }
357 found_level = btrfs_header_level(eb);
358
359 csum_tree_block(root, eb, 0);
360 err:
361 free_extent_buffer(eb);
362 out:
363 return 0;
364 }
365
366 static int check_tree_block_fsid(struct btrfs_root *root,
367 struct extent_buffer *eb)
368 {
369 struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices;
370 u8 fsid[BTRFS_UUID_SIZE];
371 int ret = 1;
372
373 read_extent_buffer(eb, fsid, (unsigned long)btrfs_header_fsid(eb),
374 BTRFS_FSID_SIZE);
375 while (fs_devices) {
376 if (!memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE)) {
377 ret = 0;
378 break;
379 }
380 fs_devices = fs_devices->seed;
381 }
382 return ret;
383 }
384
385 #ifdef CONFIG_DEBUG_LOCK_ALLOC
386 void btrfs_set_buffer_lockdep_class(struct extent_buffer *eb, int level)
387 {
388 lockdep_set_class_and_name(&eb->lock,
389 &btrfs_eb_class[level],
390 btrfs_eb_name[level]);
391 }
392 #endif
393
394 static int btree_readpage_end_io_hook(struct page *page, u64 start, u64 end,
395 struct extent_state *state)
396 {
397 struct extent_io_tree *tree;
398 u64 found_start;
399 int found_level;
400 unsigned long len;
401 struct extent_buffer *eb;
402 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
403 int ret = 0;
404
405 tree = &BTRFS_I(page->mapping->host)->io_tree;
406 if (page->private == EXTENT_PAGE_PRIVATE)
407 goto out;
408 if (!page->private)
409 goto out;
410
411 len = page->private >> 2;
412 WARN_ON(len == 0);
413
414 eb = alloc_extent_buffer(tree, start, len, page, GFP_NOFS);
415
416 found_start = btrfs_header_bytenr(eb);
417 if (found_start != start) {
418 printk(KERN_INFO "btrfs bad tree block start %llu %llu\n",
419 (unsigned long long)found_start,
420 (unsigned long long)eb->start);
421 ret = -EIO;
422 goto err;
423 }
424 if (eb->first_page != page) {
425 printk(KERN_INFO "btrfs bad first page %lu %lu\n",
426 eb->first_page->index, page->index);
427 WARN_ON(1);
428 ret = -EIO;
429 goto err;
430 }
431 if (check_tree_block_fsid(root, eb)) {
432 printk(KERN_INFO "btrfs bad fsid on block %llu\n",
433 (unsigned long long)eb->start);
434 ret = -EIO;
435 goto err;
436 }
437 found_level = btrfs_header_level(eb);
438
439 btrfs_set_buffer_lockdep_class(eb, found_level);
440
441 ret = csum_tree_block(root, eb, 1);
442 if (ret)
443 ret = -EIO;
444
445 end = min_t(u64, eb->len, PAGE_CACHE_SIZE);
446 end = eb->start + end - 1;
447 err:
448 free_extent_buffer(eb);
449 out:
450 return ret;
451 }
452
453 static void end_workqueue_bio(struct bio *bio, int err)
454 {
455 struct end_io_wq *end_io_wq = bio->bi_private;
456 struct btrfs_fs_info *fs_info;
457
458 fs_info = end_io_wq->info;
459 end_io_wq->error = err;
460 end_io_wq->work.func = end_workqueue_fn;
461 end_io_wq->work.flags = 0;
462
463 if (bio->bi_rw & (1 << BIO_RW)) {
464 if (end_io_wq->metadata)
465 btrfs_queue_worker(&fs_info->endio_meta_write_workers,
466 &end_io_wq->work);
467 else
468 btrfs_queue_worker(&fs_info->endio_write_workers,
469 &end_io_wq->work);
470 } else {
471 if (end_io_wq->metadata)
472 btrfs_queue_worker(&fs_info->endio_meta_workers,
473 &end_io_wq->work);
474 else
475 btrfs_queue_worker(&fs_info->endio_workers,
476 &end_io_wq->work);
477 }
478 }
479
480 int btrfs_bio_wq_end_io(struct btrfs_fs_info *info, struct bio *bio,
481 int metadata)
482 {
483 struct end_io_wq *end_io_wq;
484 end_io_wq = kmalloc(sizeof(*end_io_wq), GFP_NOFS);
485 if (!end_io_wq)
486 return -ENOMEM;
487
488 end_io_wq->private = bio->bi_private;
489 end_io_wq->end_io = bio->bi_end_io;
490 end_io_wq->info = info;
491 end_io_wq->error = 0;
492 end_io_wq->bio = bio;
493 end_io_wq->metadata = metadata;
494
495 bio->bi_private = end_io_wq;
496 bio->bi_end_io = end_workqueue_bio;
497 return 0;
498 }
499
500 unsigned long btrfs_async_submit_limit(struct btrfs_fs_info *info)
501 {
502 unsigned long limit = min_t(unsigned long,
503 info->workers.max_workers,
504 info->fs_devices->open_devices);
505 return 256 * limit;
506 }
507
508 int btrfs_congested_async(struct btrfs_fs_info *info, int iodone)
509 {
510 return atomic_read(&info->nr_async_bios) >
511 btrfs_async_submit_limit(info);
512 }
513
514 static void run_one_async_start(struct btrfs_work *work)
515 {
516 struct btrfs_fs_info *fs_info;
517 struct async_submit_bio *async;
518
519 async = container_of(work, struct async_submit_bio, work);
520 fs_info = BTRFS_I(async->inode)->root->fs_info;
521 async->submit_bio_start(async->inode, async->rw, async->bio,
522 async->mirror_num, async->bio_flags);
523 }
524
525 static void run_one_async_done(struct btrfs_work *work)
526 {
527 struct btrfs_fs_info *fs_info;
528 struct async_submit_bio *async;
529 int limit;
530
531 async = container_of(work, struct async_submit_bio, work);
532 fs_info = BTRFS_I(async->inode)->root->fs_info;
533
534 limit = btrfs_async_submit_limit(fs_info);
535 limit = limit * 2 / 3;
536
537 atomic_dec(&fs_info->nr_async_submits);
538
539 if (atomic_read(&fs_info->nr_async_submits) < limit &&
540 waitqueue_active(&fs_info->async_submit_wait))
541 wake_up(&fs_info->async_submit_wait);
542
543 async->submit_bio_done(async->inode, async->rw, async->bio,
544 async->mirror_num, async->bio_flags);
545 }
546
547 static void run_one_async_free(struct btrfs_work *work)
548 {
549 struct async_submit_bio *async;
550
551 async = container_of(work, struct async_submit_bio, work);
552 kfree(async);
553 }
554
555 int btrfs_wq_submit_bio(struct btrfs_fs_info *fs_info, struct inode *inode,
556 int rw, struct bio *bio, int mirror_num,
557 unsigned long bio_flags,
558 extent_submit_bio_hook_t *submit_bio_start,
559 extent_submit_bio_hook_t *submit_bio_done)
560 {
561 struct async_submit_bio *async;
562
563 async = kmalloc(sizeof(*async), GFP_NOFS);
564 if (!async)
565 return -ENOMEM;
566
567 async->inode = inode;
568 async->rw = rw;
569 async->bio = bio;
570 async->mirror_num = mirror_num;
571 async->submit_bio_start = submit_bio_start;
572 async->submit_bio_done = submit_bio_done;
573
574 async->work.func = run_one_async_start;
575 async->work.ordered_func = run_one_async_done;
576 async->work.ordered_free = run_one_async_free;
577
578 async->work.flags = 0;
579 async->bio_flags = bio_flags;
580
581 atomic_inc(&fs_info->nr_async_submits);
582
583 if (rw & (1 << BIO_RW_SYNCIO))
584 btrfs_set_work_high_prio(&async->work);
585
586 btrfs_queue_worker(&fs_info->workers, &async->work);
587 #if 0
588 int limit = btrfs_async_submit_limit(fs_info);
589 if (atomic_read(&fs_info->nr_async_submits) > limit) {
590 wait_event_timeout(fs_info->async_submit_wait,
591 (atomic_read(&fs_info->nr_async_submits) < limit),
592 HZ/10);
593
594 wait_event_timeout(fs_info->async_submit_wait,
595 (atomic_read(&fs_info->nr_async_bios) < limit),
596 HZ/10);
597 }
598 #endif
599 while (atomic_read(&fs_info->async_submit_draining) &&
600 atomic_read(&fs_info->nr_async_submits)) {
601 wait_event(fs_info->async_submit_wait,
602 (atomic_read(&fs_info->nr_async_submits) == 0));
603 }
604
605 return 0;
606 }
607
608 static int btree_csum_one_bio(struct bio *bio)
609 {
610 struct bio_vec *bvec = bio->bi_io_vec;
611 int bio_index = 0;
612 struct btrfs_root *root;
613
614 WARN_ON(bio->bi_vcnt <= 0);
615 while (bio_index < bio->bi_vcnt) {
616 root = BTRFS_I(bvec->bv_page->mapping->host)->root;
617 csum_dirty_buffer(root, bvec->bv_page);
618 bio_index++;
619 bvec++;
620 }
621 return 0;
622 }
623
624 static int __btree_submit_bio_start(struct inode *inode, int rw,
625 struct bio *bio, int mirror_num,
626 unsigned long bio_flags)
627 {
628 /*
629 * when we're called for a write, we're already in the async
630 * submission context. Just jump into btrfs_map_bio
631 */
632 btree_csum_one_bio(bio);
633 return 0;
634 }
635
636 static int __btree_submit_bio_done(struct inode *inode, int rw, struct bio *bio,
637 int mirror_num, unsigned long bio_flags)
638 {
639 /*
640 * when we're called for a write, we're already in the async
641 * submission context. Just jump into btrfs_map_bio
642 */
643 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio, mirror_num, 1);
644 }
645
646 static int btree_submit_bio_hook(struct inode *inode, int rw, struct bio *bio,
647 int mirror_num, unsigned long bio_flags)
648 {
649 int ret;
650
651 ret = btrfs_bio_wq_end_io(BTRFS_I(inode)->root->fs_info,
652 bio, 1);
653 BUG_ON(ret);
654
655 if (!(rw & (1 << BIO_RW))) {
656 /*
657 * called for a read, do the setup so that checksum validation
658 * can happen in the async kernel threads
659 */
660 return btrfs_map_bio(BTRFS_I(inode)->root, rw, bio,
661 mirror_num, 0);
662 }
663
664 /*
665 * kthread helpers are used to submit writes so that checksumming
666 * can happen in parallel across all CPUs
667 */
668 return btrfs_wq_submit_bio(BTRFS_I(inode)->root->fs_info,
669 inode, rw, bio, mirror_num, 0,
670 __btree_submit_bio_start,
671 __btree_submit_bio_done);
672 }
673
674 static int btree_writepage(struct page *page, struct writeback_control *wbc)
675 {
676 struct extent_io_tree *tree;
677 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
678 struct extent_buffer *eb;
679 int was_dirty;
680
681 tree = &BTRFS_I(page->mapping->host)->io_tree;
682 if (!(current->flags & PF_MEMALLOC)) {
683 return extent_write_full_page(tree, page,
684 btree_get_extent, wbc);
685 }
686
687 redirty_page_for_writepage(wbc, page);
688 eb = btrfs_find_tree_block(root, page_offset(page),
689 PAGE_CACHE_SIZE);
690 WARN_ON(!eb);
691
692 was_dirty = test_and_set_bit(EXTENT_BUFFER_DIRTY, &eb->bflags);
693 if (!was_dirty) {
694 spin_lock(&root->fs_info->delalloc_lock);
695 root->fs_info->dirty_metadata_bytes += PAGE_CACHE_SIZE;
696 spin_unlock(&root->fs_info->delalloc_lock);
697 }
698 free_extent_buffer(eb);
699
700 unlock_page(page);
701 return 0;
702 }
703
704 static int btree_writepages(struct address_space *mapping,
705 struct writeback_control *wbc)
706 {
707 struct extent_io_tree *tree;
708 tree = &BTRFS_I(mapping->host)->io_tree;
709 if (wbc->sync_mode == WB_SYNC_NONE) {
710 struct btrfs_root *root = BTRFS_I(mapping->host)->root;
711 u64 num_dirty;
712 unsigned long thresh = 32 * 1024 * 1024;
713
714 if (wbc->for_kupdate)
715 return 0;
716
717 /* this is a bit racy, but that's ok */
718 num_dirty = root->fs_info->dirty_metadata_bytes;
719 if (num_dirty < thresh)
720 return 0;
721 }
722 return extent_writepages(tree, mapping, btree_get_extent, wbc);
723 }
724
725 static int btree_readpage(struct file *file, struct page *page)
726 {
727 struct extent_io_tree *tree;
728 tree = &BTRFS_I(page->mapping->host)->io_tree;
729 return extent_read_full_page(tree, page, btree_get_extent);
730 }
731
732 static int btree_releasepage(struct page *page, gfp_t gfp_flags)
733 {
734 struct extent_io_tree *tree;
735 struct extent_map_tree *map;
736 int ret;
737
738 if (PageWriteback(page) || PageDirty(page))
739 return 0;
740
741 tree = &BTRFS_I(page->mapping->host)->io_tree;
742 map = &BTRFS_I(page->mapping->host)->extent_tree;
743
744 ret = try_release_extent_state(map, tree, page, gfp_flags);
745 if (!ret)
746 return 0;
747
748 ret = try_release_extent_buffer(tree, page);
749 if (ret == 1) {
750 ClearPagePrivate(page);
751 set_page_private(page, 0);
752 page_cache_release(page);
753 }
754
755 return ret;
756 }
757
758 static void btree_invalidatepage(struct page *page, unsigned long offset)
759 {
760 struct extent_io_tree *tree;
761 tree = &BTRFS_I(page->mapping->host)->io_tree;
762 extent_invalidatepage(tree, page, offset);
763 btree_releasepage(page, GFP_NOFS);
764 if (PagePrivate(page)) {
765 printk(KERN_WARNING "btrfs warning page private not zero "
766 "on page %llu\n", (unsigned long long)page_offset(page));
767 ClearPagePrivate(page);
768 set_page_private(page, 0);
769 page_cache_release(page);
770 }
771 }
772
773 #if 0
774 static int btree_writepage(struct page *page, struct writeback_control *wbc)
775 {
776 struct buffer_head *bh;
777 struct btrfs_root *root = BTRFS_I(page->mapping->host)->root;
778 struct buffer_head *head;
779 if (!page_has_buffers(page)) {
780 create_empty_buffers(page, root->fs_info->sb->s_blocksize,
781 (1 << BH_Dirty)|(1 << BH_Uptodate));
782 }
783 head = page_buffers(page);
784 bh = head;
785 do {
786 if (buffer_dirty(bh))
787 csum_tree_block(root, bh, 0);
788 bh = bh->b_this_page;
789 } while (bh != head);
790 return block_write_full_page(page, btree_get_block, wbc);
791 }
792 #endif
793
794 static struct address_space_operations btree_aops = {
795 .readpage = btree_readpage,
796 .writepage = btree_writepage,
797 .writepages = btree_writepages,
798 .releasepage = btree_releasepage,
799 .invalidatepage = btree_invalidatepage,
800 .sync_page = block_sync_page,
801 };
802
803 int readahead_tree_block(struct btrfs_root *root, u64 bytenr, u32 blocksize,
804 u64 parent_transid)
805 {
806 struct extent_buffer *buf = NULL;
807 struct inode *btree_inode = root->fs_info->btree_inode;
808 int ret = 0;
809
810 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
811 if (!buf)
812 return 0;
813 read_extent_buffer_pages(&BTRFS_I(btree_inode)->io_tree,
814 buf, 0, 0, btree_get_extent, 0);
815 free_extent_buffer(buf);
816 return ret;
817 }
818
819 struct extent_buffer *btrfs_find_tree_block(struct btrfs_root *root,
820 u64 bytenr, u32 blocksize)
821 {
822 struct inode *btree_inode = root->fs_info->btree_inode;
823 struct extent_buffer *eb;
824 eb = find_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
825 bytenr, blocksize, GFP_NOFS);
826 return eb;
827 }
828
829 struct extent_buffer *btrfs_find_create_tree_block(struct btrfs_root *root,
830 u64 bytenr, u32 blocksize)
831 {
832 struct inode *btree_inode = root->fs_info->btree_inode;
833 struct extent_buffer *eb;
834
835 eb = alloc_extent_buffer(&BTRFS_I(btree_inode)->io_tree,
836 bytenr, blocksize, NULL, GFP_NOFS);
837 return eb;
838 }
839
840
841 int btrfs_write_tree_block(struct extent_buffer *buf)
842 {
843 return btrfs_fdatawrite_range(buf->first_page->mapping, buf->start,
844 buf->start + buf->len - 1, WB_SYNC_ALL);
845 }
846
847 int btrfs_wait_tree_block_writeback(struct extent_buffer *buf)
848 {
849 return btrfs_wait_on_page_writeback_range(buf->first_page->mapping,
850 buf->start, buf->start + buf->len - 1);
851 }
852
853 struct extent_buffer *read_tree_block(struct btrfs_root *root, u64 bytenr,
854 u32 blocksize, u64 parent_transid)
855 {
856 struct extent_buffer *buf = NULL;
857 struct inode *btree_inode = root->fs_info->btree_inode;
858 struct extent_io_tree *io_tree;
859 int ret;
860
861 io_tree = &BTRFS_I(btree_inode)->io_tree;
862
863 buf = btrfs_find_create_tree_block(root, bytenr, blocksize);
864 if (!buf)
865 return NULL;
866
867 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
868
869 if (ret == 0)
870 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
871 else
872 WARN_ON(1);
873 return buf;
874
875 }
876
877 int clean_tree_block(struct btrfs_trans_handle *trans, struct btrfs_root *root,
878 struct extent_buffer *buf)
879 {
880 struct inode *btree_inode = root->fs_info->btree_inode;
881 if (btrfs_header_generation(buf) ==
882 root->fs_info->running_transaction->transid) {
883 btrfs_assert_tree_locked(buf);
884
885 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &buf->bflags)) {
886 spin_lock(&root->fs_info->delalloc_lock);
887 if (root->fs_info->dirty_metadata_bytes >= buf->len)
888 root->fs_info->dirty_metadata_bytes -= buf->len;
889 else
890 WARN_ON(1);
891 spin_unlock(&root->fs_info->delalloc_lock);
892 }
893
894 /* ugh, clear_extent_buffer_dirty needs to lock the page */
895 btrfs_set_lock_blocking(buf);
896 clear_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
897 buf);
898 }
899 return 0;
900 }
901
902 static int __setup_root(u32 nodesize, u32 leafsize, u32 sectorsize,
903 u32 stripesize, struct btrfs_root *root,
904 struct btrfs_fs_info *fs_info,
905 u64 objectid)
906 {
907 root->node = NULL;
908 root->commit_root = NULL;
909 root->ref_tree = NULL;
910 root->sectorsize = sectorsize;
911 root->nodesize = nodesize;
912 root->leafsize = leafsize;
913 root->stripesize = stripesize;
914 root->ref_cows = 0;
915 root->track_dirty = 0;
916
917 root->fs_info = fs_info;
918 root->objectid = objectid;
919 root->last_trans = 0;
920 root->highest_inode = 0;
921 root->last_inode_alloc = 0;
922 root->name = NULL;
923 root->in_sysfs = 0;
924
925 INIT_LIST_HEAD(&root->dirty_list);
926 INIT_LIST_HEAD(&root->orphan_list);
927 INIT_LIST_HEAD(&root->dead_list);
928 spin_lock_init(&root->node_lock);
929 spin_lock_init(&root->list_lock);
930 mutex_init(&root->objectid_mutex);
931 mutex_init(&root->log_mutex);
932 init_waitqueue_head(&root->log_writer_wait);
933 init_waitqueue_head(&root->log_commit_wait[0]);
934 init_waitqueue_head(&root->log_commit_wait[1]);
935 atomic_set(&root->log_commit[0], 0);
936 atomic_set(&root->log_commit[1], 0);
937 atomic_set(&root->log_writers, 0);
938 root->log_batch = 0;
939 root->log_transid = 0;
940 extent_io_tree_init(&root->dirty_log_pages,
941 fs_info->btree_inode->i_mapping, GFP_NOFS);
942
943 btrfs_leaf_ref_tree_init(&root->ref_tree_struct);
944 root->ref_tree = &root->ref_tree_struct;
945
946 memset(&root->root_key, 0, sizeof(root->root_key));
947 memset(&root->root_item, 0, sizeof(root->root_item));
948 memset(&root->defrag_progress, 0, sizeof(root->defrag_progress));
949 memset(&root->root_kobj, 0, sizeof(root->root_kobj));
950 root->defrag_trans_start = fs_info->generation;
951 init_completion(&root->kobj_unregister);
952 root->defrag_running = 0;
953 root->defrag_level = 0;
954 root->root_key.objectid = objectid;
955 root->anon_super.s_root = NULL;
956 root->anon_super.s_dev = 0;
957 INIT_LIST_HEAD(&root->anon_super.s_list);
958 INIT_LIST_HEAD(&root->anon_super.s_instances);
959 init_rwsem(&root->anon_super.s_umount);
960
961 return 0;
962 }
963
964 static int find_and_setup_root(struct btrfs_root *tree_root,
965 struct btrfs_fs_info *fs_info,
966 u64 objectid,
967 struct btrfs_root *root)
968 {
969 int ret;
970 u32 blocksize;
971 u64 generation;
972
973 __setup_root(tree_root->nodesize, tree_root->leafsize,
974 tree_root->sectorsize, tree_root->stripesize,
975 root, fs_info, objectid);
976 ret = btrfs_find_last_root(tree_root, objectid,
977 &root->root_item, &root->root_key);
978 BUG_ON(ret);
979
980 generation = btrfs_root_generation(&root->root_item);
981 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
982 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
983 blocksize, generation);
984 BUG_ON(!root->node);
985 return 0;
986 }
987
988 int btrfs_free_log_root_tree(struct btrfs_trans_handle *trans,
989 struct btrfs_fs_info *fs_info)
990 {
991 struct extent_buffer *eb;
992 struct btrfs_root *log_root_tree = fs_info->log_root_tree;
993 u64 start = 0;
994 u64 end = 0;
995 int ret;
996
997 if (!log_root_tree)
998 return 0;
999
1000 while (1) {
1001 ret = find_first_extent_bit(&log_root_tree->dirty_log_pages,
1002 0, &start, &end, EXTENT_DIRTY);
1003 if (ret)
1004 break;
1005
1006 clear_extent_dirty(&log_root_tree->dirty_log_pages,
1007 start, end, GFP_NOFS);
1008 }
1009 eb = fs_info->log_root_tree->node;
1010
1011 WARN_ON(btrfs_header_level(eb) != 0);
1012 WARN_ON(btrfs_header_nritems(eb) != 0);
1013
1014 ret = btrfs_free_reserved_extent(fs_info->tree_root,
1015 eb->start, eb->len);
1016 BUG_ON(ret);
1017
1018 free_extent_buffer(eb);
1019 kfree(fs_info->log_root_tree);
1020 fs_info->log_root_tree = NULL;
1021 return 0;
1022 }
1023
1024 static struct btrfs_root *alloc_log_tree(struct btrfs_trans_handle *trans,
1025 struct btrfs_fs_info *fs_info)
1026 {
1027 struct btrfs_root *root;
1028 struct btrfs_root *tree_root = fs_info->tree_root;
1029 struct extent_buffer *leaf;
1030
1031 root = kzalloc(sizeof(*root), GFP_NOFS);
1032 if (!root)
1033 return ERR_PTR(-ENOMEM);
1034
1035 __setup_root(tree_root->nodesize, tree_root->leafsize,
1036 tree_root->sectorsize, tree_root->stripesize,
1037 root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1038
1039 root->root_key.objectid = BTRFS_TREE_LOG_OBJECTID;
1040 root->root_key.type = BTRFS_ROOT_ITEM_KEY;
1041 root->root_key.offset = BTRFS_TREE_LOG_OBJECTID;
1042 /*
1043 * log trees do not get reference counted because they go away
1044 * before a real commit is actually done. They do store pointers
1045 * to file data extents, and those reference counts still get
1046 * updated (along with back refs to the log tree).
1047 */
1048 root->ref_cows = 0;
1049
1050 leaf = btrfs_alloc_free_block(trans, root, root->leafsize,
1051 0, BTRFS_TREE_LOG_OBJECTID,
1052 trans->transid, 0, 0, 0);
1053 if (IS_ERR(leaf)) {
1054 kfree(root);
1055 return ERR_CAST(leaf);
1056 }
1057
1058 root->node = leaf;
1059 btrfs_set_header_nritems(root->node, 0);
1060 btrfs_set_header_level(root->node, 0);
1061 btrfs_set_header_bytenr(root->node, root->node->start);
1062 btrfs_set_header_generation(root->node, trans->transid);
1063 btrfs_set_header_owner(root->node, BTRFS_TREE_LOG_OBJECTID);
1064
1065 write_extent_buffer(root->node, root->fs_info->fsid,
1066 (unsigned long)btrfs_header_fsid(root->node),
1067 BTRFS_FSID_SIZE);
1068 btrfs_mark_buffer_dirty(root->node);
1069 btrfs_tree_unlock(root->node);
1070 return root;
1071 }
1072
1073 int btrfs_init_log_root_tree(struct btrfs_trans_handle *trans,
1074 struct btrfs_fs_info *fs_info)
1075 {
1076 struct btrfs_root *log_root;
1077
1078 log_root = alloc_log_tree(trans, fs_info);
1079 if (IS_ERR(log_root))
1080 return PTR_ERR(log_root);
1081 WARN_ON(fs_info->log_root_tree);
1082 fs_info->log_root_tree = log_root;
1083 return 0;
1084 }
1085
1086 int btrfs_add_log_tree(struct btrfs_trans_handle *trans,
1087 struct btrfs_root *root)
1088 {
1089 struct btrfs_root *log_root;
1090 struct btrfs_inode_item *inode_item;
1091
1092 log_root = alloc_log_tree(trans, root->fs_info);
1093 if (IS_ERR(log_root))
1094 return PTR_ERR(log_root);
1095
1096 log_root->last_trans = trans->transid;
1097 log_root->root_key.offset = root->root_key.objectid;
1098
1099 inode_item = &log_root->root_item.inode;
1100 inode_item->generation = cpu_to_le64(1);
1101 inode_item->size = cpu_to_le64(3);
1102 inode_item->nlink = cpu_to_le32(1);
1103 inode_item->nbytes = cpu_to_le64(root->leafsize);
1104 inode_item->mode = cpu_to_le32(S_IFDIR | 0755);
1105
1106 btrfs_set_root_bytenr(&log_root->root_item, log_root->node->start);
1107 btrfs_set_root_generation(&log_root->root_item, trans->transid);
1108
1109 WARN_ON(root->log_root);
1110 root->log_root = log_root;
1111 root->log_transid = 0;
1112 return 0;
1113 }
1114
1115 struct btrfs_root *btrfs_read_fs_root_no_radix(struct btrfs_root *tree_root,
1116 struct btrfs_key *location)
1117 {
1118 struct btrfs_root *root;
1119 struct btrfs_fs_info *fs_info = tree_root->fs_info;
1120 struct btrfs_path *path;
1121 struct extent_buffer *l;
1122 u64 highest_inode;
1123 u64 generation;
1124 u32 blocksize;
1125 int ret = 0;
1126
1127 root = kzalloc(sizeof(*root), GFP_NOFS);
1128 if (!root)
1129 return ERR_PTR(-ENOMEM);
1130 if (location->offset == (u64)-1) {
1131 ret = find_and_setup_root(tree_root, fs_info,
1132 location->objectid, root);
1133 if (ret) {
1134 kfree(root);
1135 return ERR_PTR(ret);
1136 }
1137 goto insert;
1138 }
1139
1140 __setup_root(tree_root->nodesize, tree_root->leafsize,
1141 tree_root->sectorsize, tree_root->stripesize,
1142 root, fs_info, location->objectid);
1143
1144 path = btrfs_alloc_path();
1145 BUG_ON(!path);
1146 ret = btrfs_search_slot(NULL, tree_root, location, path, 0, 0);
1147 if (ret != 0) {
1148 if (ret > 0)
1149 ret = -ENOENT;
1150 goto out;
1151 }
1152 l = path->nodes[0];
1153 read_extent_buffer(l, &root->root_item,
1154 btrfs_item_ptr_offset(l, path->slots[0]),
1155 sizeof(root->root_item));
1156 memcpy(&root->root_key, location, sizeof(*location));
1157 ret = 0;
1158 out:
1159 btrfs_release_path(root, path);
1160 btrfs_free_path(path);
1161 if (ret) {
1162 kfree(root);
1163 return ERR_PTR(ret);
1164 }
1165 generation = btrfs_root_generation(&root->root_item);
1166 blocksize = btrfs_level_size(root, btrfs_root_level(&root->root_item));
1167 root->node = read_tree_block(root, btrfs_root_bytenr(&root->root_item),
1168 blocksize, generation);
1169 BUG_ON(!root->node);
1170 insert:
1171 if (location->objectid != BTRFS_TREE_LOG_OBJECTID) {
1172 root->ref_cows = 1;
1173 ret = btrfs_find_highest_inode(root, &highest_inode);
1174 if (ret == 0) {
1175 root->highest_inode = highest_inode;
1176 root->last_inode_alloc = highest_inode;
1177 }
1178 }
1179 return root;
1180 }
1181
1182 struct btrfs_root *btrfs_lookup_fs_root(struct btrfs_fs_info *fs_info,
1183 u64 root_objectid)
1184 {
1185 struct btrfs_root *root;
1186
1187 if (root_objectid == BTRFS_ROOT_TREE_OBJECTID)
1188 return fs_info->tree_root;
1189 if (root_objectid == BTRFS_EXTENT_TREE_OBJECTID)
1190 return fs_info->extent_root;
1191
1192 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1193 (unsigned long)root_objectid);
1194 return root;
1195 }
1196
1197 struct btrfs_root *btrfs_read_fs_root_no_name(struct btrfs_fs_info *fs_info,
1198 struct btrfs_key *location)
1199 {
1200 struct btrfs_root *root;
1201 int ret;
1202
1203 if (location->objectid == BTRFS_ROOT_TREE_OBJECTID)
1204 return fs_info->tree_root;
1205 if (location->objectid == BTRFS_EXTENT_TREE_OBJECTID)
1206 return fs_info->extent_root;
1207 if (location->objectid == BTRFS_CHUNK_TREE_OBJECTID)
1208 return fs_info->chunk_root;
1209 if (location->objectid == BTRFS_DEV_TREE_OBJECTID)
1210 return fs_info->dev_root;
1211 if (location->objectid == BTRFS_CSUM_TREE_OBJECTID)
1212 return fs_info->csum_root;
1213
1214 root = radix_tree_lookup(&fs_info->fs_roots_radix,
1215 (unsigned long)location->objectid);
1216 if (root)
1217 return root;
1218
1219 root = btrfs_read_fs_root_no_radix(fs_info->tree_root, location);
1220 if (IS_ERR(root))
1221 return root;
1222
1223 set_anon_super(&root->anon_super, NULL);
1224
1225 ret = radix_tree_insert(&fs_info->fs_roots_radix,
1226 (unsigned long)root->root_key.objectid,
1227 root);
1228 if (ret) {
1229 free_extent_buffer(root->node);
1230 kfree(root);
1231 return ERR_PTR(ret);
1232 }
1233 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
1234 ret = btrfs_find_dead_roots(fs_info->tree_root,
1235 root->root_key.objectid, root);
1236 BUG_ON(ret);
1237 btrfs_orphan_cleanup(root);
1238 }
1239 return root;
1240 }
1241
1242 struct btrfs_root *btrfs_read_fs_root(struct btrfs_fs_info *fs_info,
1243 struct btrfs_key *location,
1244 const char *name, int namelen)
1245 {
1246 struct btrfs_root *root;
1247 int ret;
1248
1249 root = btrfs_read_fs_root_no_name(fs_info, location);
1250 if (!root)
1251 return NULL;
1252
1253 if (root->in_sysfs)
1254 return root;
1255
1256 ret = btrfs_set_root_name(root, name, namelen);
1257 if (ret) {
1258 free_extent_buffer(root->node);
1259 kfree(root);
1260 return ERR_PTR(ret);
1261 }
1262 #if 0
1263 ret = btrfs_sysfs_add_root(root);
1264 if (ret) {
1265 free_extent_buffer(root->node);
1266 kfree(root->name);
1267 kfree(root);
1268 return ERR_PTR(ret);
1269 }
1270 #endif
1271 root->in_sysfs = 1;
1272 return root;
1273 }
1274
1275 static int btrfs_congested_fn(void *congested_data, int bdi_bits)
1276 {
1277 struct btrfs_fs_info *info = (struct btrfs_fs_info *)congested_data;
1278 int ret = 0;
1279 struct btrfs_device *device;
1280 struct backing_dev_info *bdi;
1281 #if 0
1282 if ((bdi_bits & (1 << BDI_write_congested)) &&
1283 btrfs_congested_async(info, 0))
1284 return 1;
1285 #endif
1286 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1287 if (!device->bdev)
1288 continue;
1289 bdi = blk_get_backing_dev_info(device->bdev);
1290 if (bdi && bdi_congested(bdi, bdi_bits)) {
1291 ret = 1;
1292 break;
1293 }
1294 }
1295 return ret;
1296 }
1297
1298 /*
1299 * this unplugs every device on the box, and it is only used when page
1300 * is null
1301 */
1302 static void __unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1303 {
1304 struct btrfs_device *device;
1305 struct btrfs_fs_info *info;
1306
1307 info = (struct btrfs_fs_info *)bdi->unplug_io_data;
1308 list_for_each_entry(device, &info->fs_devices->devices, dev_list) {
1309 if (!device->bdev)
1310 continue;
1311
1312 bdi = blk_get_backing_dev_info(device->bdev);
1313 if (bdi->unplug_io_fn)
1314 bdi->unplug_io_fn(bdi, page);
1315 }
1316 }
1317
1318 static void btrfs_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
1319 {
1320 struct inode *inode;
1321 struct extent_map_tree *em_tree;
1322 struct extent_map *em;
1323 struct address_space *mapping;
1324 u64 offset;
1325
1326 /* the generic O_DIRECT read code does this */
1327 if (1 || !page) {
1328 __unplug_io_fn(bdi, page);
1329 return;
1330 }
1331
1332 /*
1333 * page->mapping may change at any time. Get a consistent copy
1334 * and use that for everything below
1335 */
1336 smp_mb();
1337 mapping = page->mapping;
1338 if (!mapping)
1339 return;
1340
1341 inode = mapping->host;
1342
1343 /*
1344 * don't do the expensive searching for a small number of
1345 * devices
1346 */
1347 if (BTRFS_I(inode)->root->fs_info->fs_devices->open_devices <= 2) {
1348 __unplug_io_fn(bdi, page);
1349 return;
1350 }
1351
1352 offset = page_offset(page);
1353
1354 em_tree = &BTRFS_I(inode)->extent_tree;
1355 spin_lock(&em_tree->lock);
1356 em = lookup_extent_mapping(em_tree, offset, PAGE_CACHE_SIZE);
1357 spin_unlock(&em_tree->lock);
1358 if (!em) {
1359 __unplug_io_fn(bdi, page);
1360 return;
1361 }
1362
1363 if (em->block_start >= EXTENT_MAP_LAST_BYTE) {
1364 free_extent_map(em);
1365 __unplug_io_fn(bdi, page);
1366 return;
1367 }
1368 offset = offset - em->start;
1369 btrfs_unplug_page(&BTRFS_I(inode)->root->fs_info->mapping_tree,
1370 em->block_start + offset, page);
1371 free_extent_map(em);
1372 }
1373
1374 static int setup_bdi(struct btrfs_fs_info *info, struct backing_dev_info *bdi)
1375 {
1376 bdi_init(bdi);
1377 bdi->ra_pages = default_backing_dev_info.ra_pages;
1378 bdi->state = 0;
1379 bdi->capabilities = default_backing_dev_info.capabilities;
1380 bdi->unplug_io_fn = btrfs_unplug_io_fn;
1381 bdi->unplug_io_data = info;
1382 bdi->congested_fn = btrfs_congested_fn;
1383 bdi->congested_data = info;
1384 return 0;
1385 }
1386
1387 static int bio_ready_for_csum(struct bio *bio)
1388 {
1389 u64 length = 0;
1390 u64 buf_len = 0;
1391 u64 start = 0;
1392 struct page *page;
1393 struct extent_io_tree *io_tree = NULL;
1394 struct btrfs_fs_info *info = NULL;
1395 struct bio_vec *bvec;
1396 int i;
1397 int ret;
1398
1399 bio_for_each_segment(bvec, bio, i) {
1400 page = bvec->bv_page;
1401 if (page->private == EXTENT_PAGE_PRIVATE) {
1402 length += bvec->bv_len;
1403 continue;
1404 }
1405 if (!page->private) {
1406 length += bvec->bv_len;
1407 continue;
1408 }
1409 length = bvec->bv_len;
1410 buf_len = page->private >> 2;
1411 start = page_offset(page) + bvec->bv_offset;
1412 io_tree = &BTRFS_I(page->mapping->host)->io_tree;
1413 info = BTRFS_I(page->mapping->host)->root->fs_info;
1414 }
1415 /* are we fully contained in this bio? */
1416 if (buf_len <= length)
1417 return 1;
1418
1419 ret = extent_range_uptodate(io_tree, start + length,
1420 start + buf_len - 1);
1421 return ret;
1422 }
1423
1424 /*
1425 * called by the kthread helper functions to finally call the bio end_io
1426 * functions. This is where read checksum verification actually happens
1427 */
1428 static void end_workqueue_fn(struct btrfs_work *work)
1429 {
1430 struct bio *bio;
1431 struct end_io_wq *end_io_wq;
1432 struct btrfs_fs_info *fs_info;
1433 int error;
1434
1435 end_io_wq = container_of(work, struct end_io_wq, work);
1436 bio = end_io_wq->bio;
1437 fs_info = end_io_wq->info;
1438
1439 /* metadata bio reads are special because the whole tree block must
1440 * be checksummed at once. This makes sure the entire block is in
1441 * ram and up to date before trying to verify things. For
1442 * blocksize <= pagesize, it is basically a noop
1443 */
1444 if (!(bio->bi_rw & (1 << BIO_RW)) && end_io_wq->metadata &&
1445 !bio_ready_for_csum(bio)) {
1446 btrfs_queue_worker(&fs_info->endio_meta_workers,
1447 &end_io_wq->work);
1448 return;
1449 }
1450 error = end_io_wq->error;
1451 bio->bi_private = end_io_wq->private;
1452 bio->bi_end_io = end_io_wq->end_io;
1453 kfree(end_io_wq);
1454 bio_endio(bio, error);
1455 }
1456
1457 static int cleaner_kthread(void *arg)
1458 {
1459 struct btrfs_root *root = arg;
1460
1461 do {
1462 smp_mb();
1463 if (root->fs_info->closing)
1464 break;
1465
1466 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1467 mutex_lock(&root->fs_info->cleaner_mutex);
1468 btrfs_clean_old_snapshots(root);
1469 mutex_unlock(&root->fs_info->cleaner_mutex);
1470
1471 if (freezing(current)) {
1472 refrigerator();
1473 } else {
1474 smp_mb();
1475 if (root->fs_info->closing)
1476 break;
1477 set_current_state(TASK_INTERRUPTIBLE);
1478 schedule();
1479 __set_current_state(TASK_RUNNING);
1480 }
1481 } while (!kthread_should_stop());
1482 return 0;
1483 }
1484
1485 static int transaction_kthread(void *arg)
1486 {
1487 struct btrfs_root *root = arg;
1488 struct btrfs_trans_handle *trans;
1489 struct btrfs_transaction *cur;
1490 unsigned long now;
1491 unsigned long delay;
1492 int ret;
1493
1494 do {
1495 smp_mb();
1496 if (root->fs_info->closing)
1497 break;
1498
1499 delay = HZ * 30;
1500 vfs_check_frozen(root->fs_info->sb, SB_FREEZE_WRITE);
1501 mutex_lock(&root->fs_info->transaction_kthread_mutex);
1502
1503 mutex_lock(&root->fs_info->trans_mutex);
1504 cur = root->fs_info->running_transaction;
1505 if (!cur) {
1506 mutex_unlock(&root->fs_info->trans_mutex);
1507 goto sleep;
1508 }
1509
1510 now = get_seconds();
1511 if (now < cur->start_time || now - cur->start_time < 30) {
1512 mutex_unlock(&root->fs_info->trans_mutex);
1513 delay = HZ * 5;
1514 goto sleep;
1515 }
1516 mutex_unlock(&root->fs_info->trans_mutex);
1517 trans = btrfs_start_transaction(root, 1);
1518 ret = btrfs_commit_transaction(trans, root);
1519
1520 sleep:
1521 wake_up_process(root->fs_info->cleaner_kthread);
1522 mutex_unlock(&root->fs_info->transaction_kthread_mutex);
1523
1524 if (freezing(current)) {
1525 refrigerator();
1526 } else {
1527 if (root->fs_info->closing)
1528 break;
1529 set_current_state(TASK_INTERRUPTIBLE);
1530 schedule_timeout(delay);
1531 __set_current_state(TASK_RUNNING);
1532 }
1533 } while (!kthread_should_stop());
1534 return 0;
1535 }
1536
1537 struct btrfs_root *open_ctree(struct super_block *sb,
1538 struct btrfs_fs_devices *fs_devices,
1539 char *options)
1540 {
1541 u32 sectorsize;
1542 u32 nodesize;
1543 u32 leafsize;
1544 u32 blocksize;
1545 u32 stripesize;
1546 u64 generation;
1547 u64 features;
1548 struct btrfs_key location;
1549 struct buffer_head *bh;
1550 struct btrfs_root *extent_root = kzalloc(sizeof(struct btrfs_root),
1551 GFP_NOFS);
1552 struct btrfs_root *csum_root = kzalloc(sizeof(struct btrfs_root),
1553 GFP_NOFS);
1554 struct btrfs_root *tree_root = kzalloc(sizeof(struct btrfs_root),
1555 GFP_NOFS);
1556 struct btrfs_fs_info *fs_info = kzalloc(sizeof(*fs_info),
1557 GFP_NOFS);
1558 struct btrfs_root *chunk_root = kzalloc(sizeof(struct btrfs_root),
1559 GFP_NOFS);
1560 struct btrfs_root *dev_root = kzalloc(sizeof(struct btrfs_root),
1561 GFP_NOFS);
1562 struct btrfs_root *log_tree_root;
1563
1564 int ret;
1565 int err = -EINVAL;
1566
1567 struct btrfs_super_block *disk_super;
1568
1569 if (!extent_root || !tree_root || !fs_info ||
1570 !chunk_root || !dev_root || !csum_root) {
1571 err = -ENOMEM;
1572 goto fail;
1573 }
1574 INIT_RADIX_TREE(&fs_info->fs_roots_radix, GFP_NOFS);
1575 INIT_LIST_HEAD(&fs_info->trans_list);
1576 INIT_LIST_HEAD(&fs_info->dead_roots);
1577 INIT_LIST_HEAD(&fs_info->hashers);
1578 INIT_LIST_HEAD(&fs_info->delalloc_inodes);
1579 INIT_LIST_HEAD(&fs_info->ordered_operations);
1580 spin_lock_init(&fs_info->delalloc_lock);
1581 spin_lock_init(&fs_info->new_trans_lock);
1582 spin_lock_init(&fs_info->ref_cache_lock);
1583
1584 init_completion(&fs_info->kobj_unregister);
1585 fs_info->tree_root = tree_root;
1586 fs_info->extent_root = extent_root;
1587 fs_info->csum_root = csum_root;
1588 fs_info->chunk_root = chunk_root;
1589 fs_info->dev_root = dev_root;
1590 fs_info->fs_devices = fs_devices;
1591 INIT_LIST_HEAD(&fs_info->dirty_cowonly_roots);
1592 INIT_LIST_HEAD(&fs_info->space_info);
1593 btrfs_mapping_init(&fs_info->mapping_tree);
1594 atomic_set(&fs_info->nr_async_submits, 0);
1595 atomic_set(&fs_info->async_delalloc_pages, 0);
1596 atomic_set(&fs_info->async_submit_draining, 0);
1597 atomic_set(&fs_info->nr_async_bios, 0);
1598 atomic_set(&fs_info->throttles, 0);
1599 atomic_set(&fs_info->throttle_gen, 0);
1600 fs_info->sb = sb;
1601 fs_info->max_extent = (u64)-1;
1602 fs_info->max_inline = 8192 * 1024;
1603 setup_bdi(fs_info, &fs_info->bdi);
1604 fs_info->btree_inode = new_inode(sb);
1605 fs_info->btree_inode->i_ino = 1;
1606 fs_info->btree_inode->i_nlink = 1;
1607
1608 fs_info->thread_pool_size = min_t(unsigned long,
1609 num_online_cpus() + 2, 8);
1610
1611 INIT_LIST_HEAD(&fs_info->ordered_extents);
1612 spin_lock_init(&fs_info->ordered_extent_lock);
1613
1614 sb->s_blocksize = 4096;
1615 sb->s_blocksize_bits = blksize_bits(4096);
1616
1617 /*
1618 * we set the i_size on the btree inode to the max possible int.
1619 * the real end of the address space is determined by all of
1620 * the devices in the system
1621 */
1622 fs_info->btree_inode->i_size = OFFSET_MAX;
1623 fs_info->btree_inode->i_mapping->a_ops = &btree_aops;
1624 fs_info->btree_inode->i_mapping->backing_dev_info = &fs_info->bdi;
1625
1626 extent_io_tree_init(&BTRFS_I(fs_info->btree_inode)->io_tree,
1627 fs_info->btree_inode->i_mapping,
1628 GFP_NOFS);
1629 extent_map_tree_init(&BTRFS_I(fs_info->btree_inode)->extent_tree,
1630 GFP_NOFS);
1631
1632 BTRFS_I(fs_info->btree_inode)->io_tree.ops = &btree_extent_io_ops;
1633
1634 spin_lock_init(&fs_info->block_group_cache_lock);
1635 fs_info->block_group_cache_tree.rb_node = NULL;
1636
1637 extent_io_tree_init(&fs_info->pinned_extents,
1638 fs_info->btree_inode->i_mapping, GFP_NOFS);
1639 fs_info->do_barriers = 1;
1640
1641 INIT_LIST_HEAD(&fs_info->dead_reloc_roots);
1642 btrfs_leaf_ref_tree_init(&fs_info->reloc_ref_tree);
1643 btrfs_leaf_ref_tree_init(&fs_info->shared_ref_tree);
1644
1645 BTRFS_I(fs_info->btree_inode)->root = tree_root;
1646 memset(&BTRFS_I(fs_info->btree_inode)->location, 0,
1647 sizeof(struct btrfs_key));
1648 insert_inode_hash(fs_info->btree_inode);
1649
1650 mutex_init(&fs_info->trans_mutex);
1651 mutex_init(&fs_info->ordered_operations_mutex);
1652 mutex_init(&fs_info->tree_log_mutex);
1653 mutex_init(&fs_info->drop_mutex);
1654 mutex_init(&fs_info->chunk_mutex);
1655 mutex_init(&fs_info->transaction_kthread_mutex);
1656 mutex_init(&fs_info->cleaner_mutex);
1657 mutex_init(&fs_info->volume_mutex);
1658 mutex_init(&fs_info->tree_reloc_mutex);
1659
1660 btrfs_init_free_cluster(&fs_info->meta_alloc_cluster);
1661 btrfs_init_free_cluster(&fs_info->data_alloc_cluster);
1662
1663 init_waitqueue_head(&fs_info->transaction_throttle);
1664 init_waitqueue_head(&fs_info->transaction_wait);
1665 init_waitqueue_head(&fs_info->async_submit_wait);
1666
1667 __setup_root(4096, 4096, 4096, 4096, tree_root,
1668 fs_info, BTRFS_ROOT_TREE_OBJECTID);
1669
1670
1671 bh = btrfs_read_dev_super(fs_devices->latest_bdev);
1672 if (!bh)
1673 goto fail_iput;
1674
1675 memcpy(&fs_info->super_copy, bh->b_data, sizeof(fs_info->super_copy));
1676 memcpy(&fs_info->super_for_commit, &fs_info->super_copy,
1677 sizeof(fs_info->super_for_commit));
1678 brelse(bh);
1679
1680 memcpy(fs_info->fsid, fs_info->super_copy.fsid, BTRFS_FSID_SIZE);
1681
1682 disk_super = &fs_info->super_copy;
1683 if (!btrfs_super_root(disk_super))
1684 goto fail_iput;
1685
1686 ret = btrfs_parse_options(tree_root, options);
1687 if (ret) {
1688 err = ret;
1689 goto fail_iput;
1690 }
1691
1692 features = btrfs_super_incompat_flags(disk_super) &
1693 ~BTRFS_FEATURE_INCOMPAT_SUPP;
1694 if (features) {
1695 printk(KERN_ERR "BTRFS: couldn't mount because of "
1696 "unsupported optional features (%Lx).\n",
1697 features);
1698 err = -EINVAL;
1699 goto fail_iput;
1700 }
1701
1702 features = btrfs_super_compat_ro_flags(disk_super) &
1703 ~BTRFS_FEATURE_COMPAT_RO_SUPP;
1704 if (!(sb->s_flags & MS_RDONLY) && features) {
1705 printk(KERN_ERR "BTRFS: couldn't mount RDWR because of "
1706 "unsupported option features (%Lx).\n",
1707 features);
1708 err = -EINVAL;
1709 goto fail_iput;
1710 }
1711
1712 /*
1713 * we need to start all the end_io workers up front because the
1714 * queue work function gets called at interrupt time, and so it
1715 * cannot dynamically grow.
1716 */
1717 btrfs_init_workers(&fs_info->workers, "worker",
1718 fs_info->thread_pool_size);
1719
1720 btrfs_init_workers(&fs_info->delalloc_workers, "delalloc",
1721 fs_info->thread_pool_size);
1722
1723 btrfs_init_workers(&fs_info->submit_workers, "submit",
1724 min_t(u64, fs_devices->num_devices,
1725 fs_info->thread_pool_size));
1726
1727 /* a higher idle thresh on the submit workers makes it much more
1728 * likely that bios will be send down in a sane order to the
1729 * devices
1730 */
1731 fs_info->submit_workers.idle_thresh = 64;
1732
1733 fs_info->workers.idle_thresh = 16;
1734 fs_info->workers.ordered = 1;
1735
1736 fs_info->delalloc_workers.idle_thresh = 2;
1737 fs_info->delalloc_workers.ordered = 1;
1738
1739 btrfs_init_workers(&fs_info->fixup_workers, "fixup", 1);
1740 btrfs_init_workers(&fs_info->endio_workers, "endio",
1741 fs_info->thread_pool_size);
1742 btrfs_init_workers(&fs_info->endio_meta_workers, "endio-meta",
1743 fs_info->thread_pool_size);
1744 btrfs_init_workers(&fs_info->endio_meta_write_workers,
1745 "endio-meta-write", fs_info->thread_pool_size);
1746 btrfs_init_workers(&fs_info->endio_write_workers, "endio-write",
1747 fs_info->thread_pool_size);
1748
1749 /*
1750 * endios are largely parallel and should have a very
1751 * low idle thresh
1752 */
1753 fs_info->endio_workers.idle_thresh = 4;
1754 fs_info->endio_meta_workers.idle_thresh = 4;
1755
1756 fs_info->endio_write_workers.idle_thresh = 64;
1757 fs_info->endio_meta_write_workers.idle_thresh = 64;
1758
1759 btrfs_start_workers(&fs_info->workers, 1);
1760 btrfs_start_workers(&fs_info->submit_workers, 1);
1761 btrfs_start_workers(&fs_info->delalloc_workers, 1);
1762 btrfs_start_workers(&fs_info->fixup_workers, 1);
1763 btrfs_start_workers(&fs_info->endio_workers, fs_info->thread_pool_size);
1764 btrfs_start_workers(&fs_info->endio_meta_workers,
1765 fs_info->thread_pool_size);
1766 btrfs_start_workers(&fs_info->endio_meta_write_workers,
1767 fs_info->thread_pool_size);
1768 btrfs_start_workers(&fs_info->endio_write_workers,
1769 fs_info->thread_pool_size);
1770
1771 fs_info->bdi.ra_pages *= btrfs_super_num_devices(disk_super);
1772 fs_info->bdi.ra_pages = max(fs_info->bdi.ra_pages,
1773 4 * 1024 * 1024 / PAGE_CACHE_SIZE);
1774
1775 nodesize = btrfs_super_nodesize(disk_super);
1776 leafsize = btrfs_super_leafsize(disk_super);
1777 sectorsize = btrfs_super_sectorsize(disk_super);
1778 stripesize = btrfs_super_stripesize(disk_super);
1779 tree_root->nodesize = nodesize;
1780 tree_root->leafsize = leafsize;
1781 tree_root->sectorsize = sectorsize;
1782 tree_root->stripesize = stripesize;
1783
1784 sb->s_blocksize = sectorsize;
1785 sb->s_blocksize_bits = blksize_bits(sectorsize);
1786
1787 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
1788 sizeof(disk_super->magic))) {
1789 printk(KERN_INFO "btrfs: valid FS not found on %s\n", sb->s_id);
1790 goto fail_sb_buffer;
1791 }
1792
1793 mutex_lock(&fs_info->chunk_mutex);
1794 ret = btrfs_read_sys_array(tree_root);
1795 mutex_unlock(&fs_info->chunk_mutex);
1796 if (ret) {
1797 printk(KERN_WARNING "btrfs: failed to read the system "
1798 "array on %s\n", sb->s_id);
1799 goto fail_sys_array;
1800 }
1801
1802 blocksize = btrfs_level_size(tree_root,
1803 btrfs_super_chunk_root_level(disk_super));
1804 generation = btrfs_super_chunk_root_generation(disk_super);
1805
1806 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1807 chunk_root, fs_info, BTRFS_CHUNK_TREE_OBJECTID);
1808
1809 chunk_root->node = read_tree_block(chunk_root,
1810 btrfs_super_chunk_root(disk_super),
1811 blocksize, generation);
1812 BUG_ON(!chunk_root->node);
1813
1814 read_extent_buffer(chunk_root->node, fs_info->chunk_tree_uuid,
1815 (unsigned long)btrfs_header_chunk_tree_uuid(chunk_root->node),
1816 BTRFS_UUID_SIZE);
1817
1818 mutex_lock(&fs_info->chunk_mutex);
1819 ret = btrfs_read_chunk_tree(chunk_root);
1820 mutex_unlock(&fs_info->chunk_mutex);
1821 if (ret) {
1822 printk(KERN_WARNING "btrfs: failed to read chunk tree on %s\n",
1823 sb->s_id);
1824 goto fail_chunk_root;
1825 }
1826
1827 btrfs_close_extra_devices(fs_devices);
1828
1829 blocksize = btrfs_level_size(tree_root,
1830 btrfs_super_root_level(disk_super));
1831 generation = btrfs_super_generation(disk_super);
1832
1833 tree_root->node = read_tree_block(tree_root,
1834 btrfs_super_root(disk_super),
1835 blocksize, generation);
1836 if (!tree_root->node)
1837 goto fail_chunk_root;
1838
1839
1840 ret = find_and_setup_root(tree_root, fs_info,
1841 BTRFS_EXTENT_TREE_OBJECTID, extent_root);
1842 if (ret)
1843 goto fail_tree_root;
1844 extent_root->track_dirty = 1;
1845
1846 ret = find_and_setup_root(tree_root, fs_info,
1847 BTRFS_DEV_TREE_OBJECTID, dev_root);
1848 dev_root->track_dirty = 1;
1849 if (ret)
1850 goto fail_extent_root;
1851
1852 ret = find_and_setup_root(tree_root, fs_info,
1853 BTRFS_CSUM_TREE_OBJECTID, csum_root);
1854 if (ret)
1855 goto fail_extent_root;
1856
1857 csum_root->track_dirty = 1;
1858
1859 btrfs_read_block_groups(extent_root);
1860
1861 fs_info->generation = generation;
1862 fs_info->last_trans_committed = generation;
1863 fs_info->data_alloc_profile = (u64)-1;
1864 fs_info->metadata_alloc_profile = (u64)-1;
1865 fs_info->system_alloc_profile = fs_info->metadata_alloc_profile;
1866 fs_info->cleaner_kthread = kthread_run(cleaner_kthread, tree_root,
1867 "btrfs-cleaner");
1868 if (IS_ERR(fs_info->cleaner_kthread))
1869 goto fail_csum_root;
1870
1871 fs_info->transaction_kthread = kthread_run(transaction_kthread,
1872 tree_root,
1873 "btrfs-transaction");
1874 if (IS_ERR(fs_info->transaction_kthread))
1875 goto fail_cleaner;
1876
1877 if (btrfs_super_log_root(disk_super) != 0) {
1878 u64 bytenr = btrfs_super_log_root(disk_super);
1879
1880 if (fs_devices->rw_devices == 0) {
1881 printk(KERN_WARNING "Btrfs log replay required "
1882 "on RO media\n");
1883 err = -EIO;
1884 goto fail_trans_kthread;
1885 }
1886 blocksize =
1887 btrfs_level_size(tree_root,
1888 btrfs_super_log_root_level(disk_super));
1889
1890 log_tree_root = kzalloc(sizeof(struct btrfs_root),
1891 GFP_NOFS);
1892
1893 __setup_root(nodesize, leafsize, sectorsize, stripesize,
1894 log_tree_root, fs_info, BTRFS_TREE_LOG_OBJECTID);
1895
1896 log_tree_root->node = read_tree_block(tree_root, bytenr,
1897 blocksize,
1898 generation + 1);
1899 ret = btrfs_recover_log_trees(log_tree_root);
1900 BUG_ON(ret);
1901
1902 if (sb->s_flags & MS_RDONLY) {
1903 ret = btrfs_commit_super(tree_root);
1904 BUG_ON(ret);
1905 }
1906 }
1907
1908 if (!(sb->s_flags & MS_RDONLY)) {
1909 ret = btrfs_cleanup_reloc_trees(tree_root);
1910 BUG_ON(ret);
1911 }
1912
1913 location.objectid = BTRFS_FS_TREE_OBJECTID;
1914 location.type = BTRFS_ROOT_ITEM_KEY;
1915 location.offset = (u64)-1;
1916
1917 fs_info->fs_root = btrfs_read_fs_root_no_name(fs_info, &location);
1918 if (!fs_info->fs_root)
1919 goto fail_trans_kthread;
1920 return tree_root;
1921
1922 fail_trans_kthread:
1923 kthread_stop(fs_info->transaction_kthread);
1924 fail_cleaner:
1925 kthread_stop(fs_info->cleaner_kthread);
1926
1927 /*
1928 * make sure we're done with the btree inode before we stop our
1929 * kthreads
1930 */
1931 filemap_write_and_wait(fs_info->btree_inode->i_mapping);
1932 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1933
1934 fail_csum_root:
1935 free_extent_buffer(csum_root->node);
1936 fail_extent_root:
1937 free_extent_buffer(extent_root->node);
1938 fail_tree_root:
1939 free_extent_buffer(tree_root->node);
1940 fail_chunk_root:
1941 free_extent_buffer(chunk_root->node);
1942 fail_sys_array:
1943 free_extent_buffer(dev_root->node);
1944 fail_sb_buffer:
1945 btrfs_stop_workers(&fs_info->fixup_workers);
1946 btrfs_stop_workers(&fs_info->delalloc_workers);
1947 btrfs_stop_workers(&fs_info->workers);
1948 btrfs_stop_workers(&fs_info->endio_workers);
1949 btrfs_stop_workers(&fs_info->endio_meta_workers);
1950 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
1951 btrfs_stop_workers(&fs_info->endio_write_workers);
1952 btrfs_stop_workers(&fs_info->submit_workers);
1953 fail_iput:
1954 invalidate_inode_pages2(fs_info->btree_inode->i_mapping);
1955 iput(fs_info->btree_inode);
1956
1957 btrfs_close_devices(fs_info->fs_devices);
1958 btrfs_mapping_tree_free(&fs_info->mapping_tree);
1959 bdi_destroy(&fs_info->bdi);
1960
1961 fail:
1962 kfree(extent_root);
1963 kfree(tree_root);
1964 kfree(fs_info);
1965 kfree(chunk_root);
1966 kfree(dev_root);
1967 kfree(csum_root);
1968 return ERR_PTR(err);
1969 }
1970
1971 static void btrfs_end_buffer_write_sync(struct buffer_head *bh, int uptodate)
1972 {
1973 char b[BDEVNAME_SIZE];
1974
1975 if (uptodate) {
1976 set_buffer_uptodate(bh);
1977 } else {
1978 if (!buffer_eopnotsupp(bh) && printk_ratelimit()) {
1979 printk(KERN_WARNING "lost page write due to "
1980 "I/O error on %s\n",
1981 bdevname(bh->b_bdev, b));
1982 }
1983 /* note, we dont' set_buffer_write_io_error because we have
1984 * our own ways of dealing with the IO errors
1985 */
1986 clear_buffer_uptodate(bh);
1987 }
1988 unlock_buffer(bh);
1989 put_bh(bh);
1990 }
1991
1992 struct buffer_head *btrfs_read_dev_super(struct block_device *bdev)
1993 {
1994 struct buffer_head *bh;
1995 struct buffer_head *latest = NULL;
1996 struct btrfs_super_block *super;
1997 int i;
1998 u64 transid = 0;
1999 u64 bytenr;
2000
2001 /* we would like to check all the supers, but that would make
2002 * a btrfs mount succeed after a mkfs from a different FS.
2003 * So, we need to add a special mount option to scan for
2004 * later supers, using BTRFS_SUPER_MIRROR_MAX instead
2005 */
2006 for (i = 0; i < 1; i++) {
2007 bytenr = btrfs_sb_offset(i);
2008 if (bytenr + 4096 >= i_size_read(bdev->bd_inode))
2009 break;
2010 bh = __bread(bdev, bytenr / 4096, 4096);
2011 if (!bh)
2012 continue;
2013
2014 super = (struct btrfs_super_block *)bh->b_data;
2015 if (btrfs_super_bytenr(super) != bytenr ||
2016 strncmp((char *)(&super->magic), BTRFS_MAGIC,
2017 sizeof(super->magic))) {
2018 brelse(bh);
2019 continue;
2020 }
2021
2022 if (!latest || btrfs_super_generation(super) > transid) {
2023 brelse(latest);
2024 latest = bh;
2025 transid = btrfs_super_generation(super);
2026 } else {
2027 brelse(bh);
2028 }
2029 }
2030 return latest;
2031 }
2032
2033 static int write_dev_supers(struct btrfs_device *device,
2034 struct btrfs_super_block *sb,
2035 int do_barriers, int wait, int max_mirrors)
2036 {
2037 struct buffer_head *bh;
2038 int i;
2039 int ret;
2040 int errors = 0;
2041 u32 crc;
2042 u64 bytenr;
2043 int last_barrier = 0;
2044
2045 if (max_mirrors == 0)
2046 max_mirrors = BTRFS_SUPER_MIRROR_MAX;
2047
2048 /* make sure only the last submit_bh does a barrier */
2049 if (do_barriers) {
2050 for (i = 0; i < max_mirrors; i++) {
2051 bytenr = btrfs_sb_offset(i);
2052 if (bytenr + BTRFS_SUPER_INFO_SIZE >=
2053 device->total_bytes)
2054 break;
2055 last_barrier = i;
2056 }
2057 }
2058
2059 for (i = 0; i < max_mirrors; i++) {
2060 bytenr = btrfs_sb_offset(i);
2061 if (bytenr + BTRFS_SUPER_INFO_SIZE >= device->total_bytes)
2062 break;
2063
2064 if (wait) {
2065 bh = __find_get_block(device->bdev, bytenr / 4096,
2066 BTRFS_SUPER_INFO_SIZE);
2067 BUG_ON(!bh);
2068 brelse(bh);
2069 wait_on_buffer(bh);
2070 if (buffer_uptodate(bh)) {
2071 brelse(bh);
2072 continue;
2073 }
2074 } else {
2075 btrfs_set_super_bytenr(sb, bytenr);
2076
2077 crc = ~(u32)0;
2078 crc = btrfs_csum_data(NULL, (char *)sb +
2079 BTRFS_CSUM_SIZE, crc,
2080 BTRFS_SUPER_INFO_SIZE -
2081 BTRFS_CSUM_SIZE);
2082 btrfs_csum_final(crc, sb->csum);
2083
2084 bh = __getblk(device->bdev, bytenr / 4096,
2085 BTRFS_SUPER_INFO_SIZE);
2086 memcpy(bh->b_data, sb, BTRFS_SUPER_INFO_SIZE);
2087
2088 set_buffer_uptodate(bh);
2089 get_bh(bh);
2090 lock_buffer(bh);
2091 bh->b_end_io = btrfs_end_buffer_write_sync;
2092 }
2093
2094 if (i == last_barrier && do_barriers && device->barriers) {
2095 ret = submit_bh(WRITE_BARRIER, bh);
2096 if (ret == -EOPNOTSUPP) {
2097 printk("btrfs: disabling barriers on dev %s\n",
2098 device->name);
2099 set_buffer_uptodate(bh);
2100 device->barriers = 0;
2101 get_bh(bh);
2102 lock_buffer(bh);
2103 ret = submit_bh(WRITE_SYNC, bh);
2104 }
2105 } else {
2106 ret = submit_bh(WRITE_SYNC, bh);
2107 }
2108
2109 if (!ret && wait) {
2110 wait_on_buffer(bh);
2111 if (!buffer_uptodate(bh))
2112 errors++;
2113 } else if (ret) {
2114 errors++;
2115 }
2116 if (wait)
2117 brelse(bh);
2118 }
2119 return errors < i ? 0 : -1;
2120 }
2121
2122 int write_all_supers(struct btrfs_root *root, int max_mirrors)
2123 {
2124 struct list_head *head = &root->fs_info->fs_devices->devices;
2125 struct btrfs_device *dev;
2126 struct btrfs_super_block *sb;
2127 struct btrfs_dev_item *dev_item;
2128 int ret;
2129 int do_barriers;
2130 int max_errors;
2131 int total_errors = 0;
2132 u64 flags;
2133
2134 max_errors = btrfs_super_num_devices(&root->fs_info->super_copy) - 1;
2135 do_barriers = !btrfs_test_opt(root, NOBARRIER);
2136
2137 sb = &root->fs_info->super_for_commit;
2138 dev_item = &sb->dev_item;
2139 list_for_each_entry(dev, head, dev_list) {
2140 if (!dev->bdev) {
2141 total_errors++;
2142 continue;
2143 }
2144 if (!dev->in_fs_metadata || !dev->writeable)
2145 continue;
2146
2147 btrfs_set_stack_device_generation(dev_item, 0);
2148 btrfs_set_stack_device_type(dev_item, dev->type);
2149 btrfs_set_stack_device_id(dev_item, dev->devid);
2150 btrfs_set_stack_device_total_bytes(dev_item, dev->total_bytes);
2151 btrfs_set_stack_device_bytes_used(dev_item, dev->bytes_used);
2152 btrfs_set_stack_device_io_align(dev_item, dev->io_align);
2153 btrfs_set_stack_device_io_width(dev_item, dev->io_width);
2154 btrfs_set_stack_device_sector_size(dev_item, dev->sector_size);
2155 memcpy(dev_item->uuid, dev->uuid, BTRFS_UUID_SIZE);
2156 memcpy(dev_item->fsid, dev->fs_devices->fsid, BTRFS_UUID_SIZE);
2157
2158 flags = btrfs_super_flags(sb);
2159 btrfs_set_super_flags(sb, flags | BTRFS_HEADER_FLAG_WRITTEN);
2160
2161 ret = write_dev_supers(dev, sb, do_barriers, 0, max_mirrors);
2162 if (ret)
2163 total_errors++;
2164 }
2165 if (total_errors > max_errors) {
2166 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2167 total_errors);
2168 BUG();
2169 }
2170
2171 total_errors = 0;
2172 list_for_each_entry(dev, head, dev_list) {
2173 if (!dev->bdev)
2174 continue;
2175 if (!dev->in_fs_metadata || !dev->writeable)
2176 continue;
2177
2178 ret = write_dev_supers(dev, sb, do_barriers, 1, max_mirrors);
2179 if (ret)
2180 total_errors++;
2181 }
2182 if (total_errors > max_errors) {
2183 printk(KERN_ERR "btrfs: %d errors while writing supers\n",
2184 total_errors);
2185 BUG();
2186 }
2187 return 0;
2188 }
2189
2190 int write_ctree_super(struct btrfs_trans_handle *trans,
2191 struct btrfs_root *root, int max_mirrors)
2192 {
2193 int ret;
2194
2195 ret = write_all_supers(root, max_mirrors);
2196 return ret;
2197 }
2198
2199 int btrfs_free_fs_root(struct btrfs_fs_info *fs_info, struct btrfs_root *root)
2200 {
2201 radix_tree_delete(&fs_info->fs_roots_radix,
2202 (unsigned long)root->root_key.objectid);
2203 if (root->anon_super.s_dev) {
2204 down_write(&root->anon_super.s_umount);
2205 kill_anon_super(&root->anon_super);
2206 }
2207 if (root->node)
2208 free_extent_buffer(root->node);
2209 if (root->commit_root)
2210 free_extent_buffer(root->commit_root);
2211 kfree(root->name);
2212 kfree(root);
2213 return 0;
2214 }
2215
2216 static int del_fs_roots(struct btrfs_fs_info *fs_info)
2217 {
2218 int ret;
2219 struct btrfs_root *gang[8];
2220 int i;
2221
2222 while (1) {
2223 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2224 (void **)gang, 0,
2225 ARRAY_SIZE(gang));
2226 if (!ret)
2227 break;
2228 for (i = 0; i < ret; i++)
2229 btrfs_free_fs_root(fs_info, gang[i]);
2230 }
2231 return 0;
2232 }
2233
2234 int btrfs_cleanup_fs_roots(struct btrfs_fs_info *fs_info)
2235 {
2236 u64 root_objectid = 0;
2237 struct btrfs_root *gang[8];
2238 int i;
2239 int ret;
2240
2241 while (1) {
2242 ret = radix_tree_gang_lookup(&fs_info->fs_roots_radix,
2243 (void **)gang, root_objectid,
2244 ARRAY_SIZE(gang));
2245 if (!ret)
2246 break;
2247 for (i = 0; i < ret; i++) {
2248 root_objectid = gang[i]->root_key.objectid;
2249 ret = btrfs_find_dead_roots(fs_info->tree_root,
2250 root_objectid, gang[i]);
2251 BUG_ON(ret);
2252 btrfs_orphan_cleanup(gang[i]);
2253 }
2254 root_objectid++;
2255 }
2256 return 0;
2257 }
2258
2259 int btrfs_commit_super(struct btrfs_root *root)
2260 {
2261 struct btrfs_trans_handle *trans;
2262 int ret;
2263
2264 mutex_lock(&root->fs_info->cleaner_mutex);
2265 btrfs_clean_old_snapshots(root);
2266 mutex_unlock(&root->fs_info->cleaner_mutex);
2267 trans = btrfs_start_transaction(root, 1);
2268 ret = btrfs_commit_transaction(trans, root);
2269 BUG_ON(ret);
2270 /* run commit again to drop the original snapshot */
2271 trans = btrfs_start_transaction(root, 1);
2272 btrfs_commit_transaction(trans, root);
2273 ret = btrfs_write_and_wait_transaction(NULL, root);
2274 BUG_ON(ret);
2275
2276 ret = write_ctree_super(NULL, root, 0);
2277 return ret;
2278 }
2279
2280 int close_ctree(struct btrfs_root *root)
2281 {
2282 struct btrfs_fs_info *fs_info = root->fs_info;
2283 int ret;
2284
2285 fs_info->closing = 1;
2286 smp_mb();
2287
2288 kthread_stop(root->fs_info->transaction_kthread);
2289 kthread_stop(root->fs_info->cleaner_kthread);
2290
2291 if (!(fs_info->sb->s_flags & MS_RDONLY)) {
2292 ret = btrfs_commit_super(root);
2293 if (ret)
2294 printk(KERN_ERR "btrfs: commit super ret %d\n", ret);
2295 }
2296
2297 if (fs_info->delalloc_bytes) {
2298 printk(KERN_INFO "btrfs: at unmount delalloc count %llu\n",
2299 fs_info->delalloc_bytes);
2300 }
2301 if (fs_info->total_ref_cache_size) {
2302 printk(KERN_INFO "btrfs: at umount reference cache size %llu\n",
2303 (unsigned long long)fs_info->total_ref_cache_size);
2304 }
2305
2306 if (fs_info->extent_root->node)
2307 free_extent_buffer(fs_info->extent_root->node);
2308
2309 if (fs_info->tree_root->node)
2310 free_extent_buffer(fs_info->tree_root->node);
2311
2312 if (root->fs_info->chunk_root->node)
2313 free_extent_buffer(root->fs_info->chunk_root->node);
2314
2315 if (root->fs_info->dev_root->node)
2316 free_extent_buffer(root->fs_info->dev_root->node);
2317
2318 if (root->fs_info->csum_root->node)
2319 free_extent_buffer(root->fs_info->csum_root->node);
2320
2321 btrfs_free_block_groups(root->fs_info);
2322
2323 del_fs_roots(fs_info);
2324
2325 iput(fs_info->btree_inode);
2326
2327 btrfs_stop_workers(&fs_info->fixup_workers);
2328 btrfs_stop_workers(&fs_info->delalloc_workers);
2329 btrfs_stop_workers(&fs_info->workers);
2330 btrfs_stop_workers(&fs_info->endio_workers);
2331 btrfs_stop_workers(&fs_info->endio_meta_workers);
2332 btrfs_stop_workers(&fs_info->endio_meta_write_workers);
2333 btrfs_stop_workers(&fs_info->endio_write_workers);
2334 btrfs_stop_workers(&fs_info->submit_workers);
2335
2336 #if 0
2337 while (!list_empty(&fs_info->hashers)) {
2338 struct btrfs_hasher *hasher;
2339 hasher = list_entry(fs_info->hashers.next, struct btrfs_hasher,
2340 hashers);
2341 list_del(&hasher->hashers);
2342 crypto_free_hash(&fs_info->hash_tfm);
2343 kfree(hasher);
2344 }
2345 #endif
2346 btrfs_close_devices(fs_info->fs_devices);
2347 btrfs_mapping_tree_free(&fs_info->mapping_tree);
2348
2349 bdi_destroy(&fs_info->bdi);
2350
2351 kfree(fs_info->extent_root);
2352 kfree(fs_info->tree_root);
2353 kfree(fs_info->chunk_root);
2354 kfree(fs_info->dev_root);
2355 kfree(fs_info->csum_root);
2356 return 0;
2357 }
2358
2359 int btrfs_buffer_uptodate(struct extent_buffer *buf, u64 parent_transid)
2360 {
2361 int ret;
2362 struct inode *btree_inode = buf->first_page->mapping->host;
2363
2364 ret = extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree, buf);
2365 if (!ret)
2366 return ret;
2367
2368 ret = verify_parent_transid(&BTRFS_I(btree_inode)->io_tree, buf,
2369 parent_transid);
2370 return !ret;
2371 }
2372
2373 int btrfs_set_buffer_uptodate(struct extent_buffer *buf)
2374 {
2375 struct inode *btree_inode = buf->first_page->mapping->host;
2376 return set_extent_buffer_uptodate(&BTRFS_I(btree_inode)->io_tree,
2377 buf);
2378 }
2379
2380 void btrfs_mark_buffer_dirty(struct extent_buffer *buf)
2381 {
2382 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2383 u64 transid = btrfs_header_generation(buf);
2384 struct inode *btree_inode = root->fs_info->btree_inode;
2385 int was_dirty;
2386
2387 btrfs_assert_tree_locked(buf);
2388 if (transid != root->fs_info->generation) {
2389 printk(KERN_CRIT "btrfs transid mismatch buffer %llu, "
2390 "found %llu running %llu\n",
2391 (unsigned long long)buf->start,
2392 (unsigned long long)transid,
2393 (unsigned long long)root->fs_info->generation);
2394 WARN_ON(1);
2395 }
2396 was_dirty = set_extent_buffer_dirty(&BTRFS_I(btree_inode)->io_tree,
2397 buf);
2398 if (!was_dirty) {
2399 spin_lock(&root->fs_info->delalloc_lock);
2400 root->fs_info->dirty_metadata_bytes += buf->len;
2401 spin_unlock(&root->fs_info->delalloc_lock);
2402 }
2403 }
2404
2405 void btrfs_btree_balance_dirty(struct btrfs_root *root, unsigned long nr)
2406 {
2407 /*
2408 * looks as though older kernels can get into trouble with
2409 * this code, they end up stuck in balance_dirty_pages forever
2410 */
2411 struct extent_io_tree *tree;
2412 u64 num_dirty;
2413 u64 start = 0;
2414 unsigned long thresh = 32 * 1024 * 1024;
2415 tree = &BTRFS_I(root->fs_info->btree_inode)->io_tree;
2416
2417 if (current->flags & PF_MEMALLOC)
2418 return;
2419
2420 num_dirty = count_range_bits(tree, &start, (u64)-1,
2421 thresh, EXTENT_DIRTY);
2422 if (num_dirty > thresh) {
2423 balance_dirty_pages_ratelimited_nr(
2424 root->fs_info->btree_inode->i_mapping, 1);
2425 }
2426 return;
2427 }
2428
2429 int btrfs_read_buffer(struct extent_buffer *buf, u64 parent_transid)
2430 {
2431 struct btrfs_root *root = BTRFS_I(buf->first_page->mapping->host)->root;
2432 int ret;
2433 ret = btree_read_extent_buffer_pages(root, buf, 0, parent_transid);
2434 if (ret == 0)
2435 set_bit(EXTENT_BUFFER_UPTODATE, &buf->bflags);
2436 return ret;
2437 }
2438
2439 int btree_lock_page_hook(struct page *page)
2440 {
2441 struct inode *inode = page->mapping->host;
2442 struct btrfs_root *root = BTRFS_I(inode)->root;
2443 struct extent_io_tree *io_tree = &BTRFS_I(inode)->io_tree;
2444 struct extent_buffer *eb;
2445 unsigned long len;
2446 u64 bytenr = page_offset(page);
2447
2448 if (page->private == EXTENT_PAGE_PRIVATE)
2449 goto out;
2450
2451 len = page->private >> 2;
2452 eb = find_extent_buffer(io_tree, bytenr, len, GFP_NOFS);
2453 if (!eb)
2454 goto out;
2455
2456 btrfs_tree_lock(eb);
2457 btrfs_set_header_flag(eb, BTRFS_HEADER_FLAG_WRITTEN);
2458
2459 if (test_and_clear_bit(EXTENT_BUFFER_DIRTY, &eb->bflags)) {
2460 spin_lock(&root->fs_info->delalloc_lock);
2461 if (root->fs_info->dirty_metadata_bytes >= eb->len)
2462 root->fs_info->dirty_metadata_bytes -= eb->len;
2463 else
2464 WARN_ON(1);
2465 spin_unlock(&root->fs_info->delalloc_lock);
2466 }
2467
2468 btrfs_tree_unlock(eb);
2469 free_extent_buffer(eb);
2470 out:
2471 lock_page(page);
2472 return 0;
2473 }
2474
2475 static struct extent_io_ops btree_extent_io_ops = {
2476 .write_cache_pages_lock_hook = btree_lock_page_hook,
2477 .readpage_end_io_hook = btree_readpage_end_io_hook,
2478 .submit_bio_hook = btree_submit_bio_hook,
2479 /* note we're sharing with inode.c for the merge bio hook */
2480 .merge_bio_hook = btrfs_merge_bio_hook,
2481 };
Ubuntu Artful kernel mirror